#asahi-dev · 2025-12-23 10:06
chadmedwebgl aquarium, glxgears, vkgears#asahi-dev · 2025-12-23 10:04
chadmedive made atomic_flush non-blocking and it helps with pacing but kwin is still sitting at 60hz until it decides not to #asahi-dev · 2025-12-23 10:03
chadmedtheres still something up with the 14/16" machines not reliably running at 120 hz. kwin still wants to sit at 60 hz but will intermittently go up to 120#asahi-dev · 2025-12-23 08:13
chadmedi havent tried explicitly setting a colourspace for our surfaces other than whatever "native" is#asahi-dev · 2025-12-23 08:06
chadmedit does this intermittently so we might not be the problem #asahi-dev · 2025-12-23 08:06
chadmedwell it should at the very minimum make my screen orange and not green :P#asahi-dev · 2025-12-23 07:45
chadmedeven with ctm-based night light on?#asahi-dev · 2025-12-23 00:33
chadmedthe colourimetry for the builtin displays is definitely wrong through and setting the right DisplayP3 profile does not fix it in the slightest #asahi-dev · 2025-12-23 00:32
chadmedi dont think we'll need one just to make dcp DTRT since it all happens in firmware and we just feed it constants to select a colour space and transfer function #asahi-dev · 2025-12-22 22:31
chadmedchanging how all of this behaves in teh driver currently is not going to be trivial though #asahi-dev · 2025-12-22 22:29
chadmednone of which we currently do#asahi-dev · 2025-12-22 22:29
chadmedwe also need to properly set the correct EOTF, colour space and transfer functions #asahi-dev · 2025-12-22 13:03
chadmedyeah thats the plan for tomorrow or after christmas. just wanted to keep as many things out of the pipeline as possible to make sure the stuff i did to the driver actually works#asahi-dev · 2025-12-22 12:48
chadmedwe have to skip vsync as kmssink relies on VBIs#asahi-dev · 2025-12-22 12:47
chadmedgst-launch-1.0 filesrc location=[path] ! qtdemux ! h264parse ! avdec_h264 ! videoconvert ! videoscale ! video/x-raw,format=NV12,width=1024,height=576 ! kmssink driver-name=apple skip-vsync=true#asahi-dev · 2025-12-22 12:43
chadmedoh right, well it is demuxing an mp4, decoding the h264 within, then converting the yuv420p to nv12 in software, then scaling that down to a sane resolution for the overlay plane #asahi-dev · 2025-12-22 12:40
chadmedjannau: no clue, but the videosrc test didnt do it #asahi-dev · 2025-12-22 12:39
chadmedleio: its not complicated its just gross :P#asahi-dev · 2025-12-22 12:19
chadmedcould just be my insane gstreamer pipeline too #asahi-dev · 2025-12-22 12:19
chadmednv12 scanout tears quite badly on both planes but does work#asahi-dev · 2025-12-22 09:12
chadmedwe havent even touched anything there in ages#asahi-dev · 2025-12-22 09:12
chadmedstarting to think this driver might be a bit of a maintenance nightmare ;)#asahi-dev · 2025-12-22 08:40
chadmedyeah looks like it. theres another call with around about the same amount of data. ive got 13.5 on my test machine now anyway#asahi-dev · 2025-12-22 07:49
chadmedA408/swap_submit_dcp disappeared with macos 14 fantastic#asahi-dev · 2025-12-22 06:21
chadmedoh they pad unswapped surfaces with 0xaa in the IOSurface array#asahi-dev · 2025-12-22 06:20
chadmedtheir surface format for their cursor fb is also super fucked up lol #asahi-dev · 2025-12-22 06:18
chadmedoverlay plane swaps never have ts1 or ts2 set, but ts3 continues ticking up#asahi-dev · 2025-12-21 23:56
chadmedfinding an installer for 13.5 is becoming incredibly difficult#asahi-dev · 2025-12-21 23:03
chadmedyeah we're just trying to understand these timestamps right now#asahi-dev · 2025-12-21 23:02
chadmedi just dont use them because im booted into linux 99.999% of the time#asahi-dev · 2025-12-21 23:01
chadmedyeah ive got dp adapters#asahi-dev · 2025-12-21 22:42
chadmedoh ill need a 13.5 install wont i#asahi-dev · 2025-12-21 22:41
chadmedyeah, ill try get macos to play nice with my vrr monitors and see what happens to the timestamps#asahi-dev · 2025-12-21 22:31
chadmedi just dont see what these three timestamps, which are so closely correlated, could possibly be other than what we have already inferred#asahi-dev · 2025-12-21 22:31
chadmedthe traces are going to be hard to interpret as meaningfully related to realtime no matter what we do because m1n1 is stealing time while our timebase still ticks up#asahi-dev · 2025-12-21 13:17
chadmed(enabling tearing in the surfaces seems to fix the laggy brightness changing and some other weird stuff btw) #asahi-dev · 2025-12-21 13:16
chadmedwould the timeout always be exactly one frame in the future though even if we're disallowing tearing#asahi-dev · 2025-12-21 13:11
chadmedi dont really see how they could be anything else given the tight timing but sure#asahi-dev · 2025-12-21 12:51
chadmedmacos doesnt like my vrr monitors though so i cant really test that with varying times between swaps for a given mode#asahi-dev · 2025-12-21 12:50
chadmedts2 is the only one im not _too_ sure is real but it seems to line up#asahi-dev · 2025-12-21 12:49
chadmedoliver, jannau: setting ts1 = now + (1 / refresh rate) * 1e9, ts2 = last time we got dcp_swapped and ts3 = now does not break anything on any of the hdmi displays ive tested#asahi-dev · 2025-12-21 07:36
chadmedthe firmware prints a "tiling unsupported" error message to its rtk syslog#asahi-dev · 2025-12-21 07:36
chadmedso dcp will happily scan out a DRM_FORMAT_MOD_APPLE_GPU_TILED{,_COMPRESSED} framebuffer but actually setting the tiled flag on the surface makes it very unhappy #asahi-dev · 2025-12-18 11:34
chadmedwe still havent really investigated timesync either right#asahi-dev · 2025-12-18 11:33
chadmedbut i would think (perhaps incorrectly) that if dcp were keeping time out of band with whatever the cpus are doing then it would keep ticking even while waiting for a swap to complete#asahi-dev · 2025-12-18 11:32
chadmedthe ADT #asahi-dev · 2025-12-18 11:31
chadmedbeen a while since but iirc macos does not like being halted for too long though#asahi-dev · 2025-12-18 11:30
chadmedor if you just want to poke around at a point in time you can break into the hypervisor on the serial console with ctrl+c and use the same functionality like an old school memory monitor#asahi-dev · 2025-12-18 11:29
chadmedhave a look at how some of the stuff in proxyclient/experiments interact with RegMonitor#asahi-dev · 2025-12-18 11:26
chadmedthat or the driver is waiting for a swap to finish then just going ts_prev + 1/refresh rate when doing the next swap